140 Chapter 5 Groep. The duration of contact between dyads was used to define the per-protocol samples (i.e., no compliance, low compliance, and high compliance). Statistical methods The statistical analyses were conducted in accordance with a prespecified analysis plan using IBM SPSS Statistics (2020), version 27.0. Descriptive statistics were used to explore baseline characteristics. Differences in baseline characteristics between the two intervention groups were examined using chi-square tests for categorical variables and independent samples t-tests or Mann-Whitney U tests for continuous variables. Given the substantial dropout of participants in the intervention, between-group differences in duration of TAU from baseline to 12-, and 18-month follow-up were examined using Mann-Whitney U tests. Missingness of data points between participants in TAU+FNC and TAU was examined using a chi-square test. Furthermore, baseline characteristics of participants with and without two or more missing data points were compared, using chi-square tests for categorical variables and independent samples t-tests or Mann-Whitney U tests for continuous variables, to explore the differences between these participant groups. Intention-to-treat analyses of the primary outcome measures and (key) secondary outcome measures were conducted, which included analyses of all the available data (between baseline and 18-month follow-up) from participants who completed baseline assessment (N = 102). In addition, per-protocol analyses were conducted on the primary (i.e., mental wellbeing) and key secondary outcomes only (i.e., general psychiatric functioning, hospitalization, criminal behavior, and incarceration), comparing different compliance groups of the FNC intervention (i.e., no compliance, low compliance, and high compliance) to the TAU intervention. Linear mixed models (LMM) analyses were used to analyze repeatedly measured continuous outcomes and normally distributed count outcomes. Generalized estimating equations (GEE) analyses were used to analyze repeatedly measured count outcomes with Poisson and negative binominal distributions, as well as one dichotomous outcome (i.e., quality of relationships with core social network members). For all LMM and GEE analyses, first the overall between-group treatment effect on average over time was analyzed and second, the between-group treatment effect at the different follow-up timepoints. For the latter, besides the intervention group variable (i.e., TAU+FNC versus TAU), also time (3-, 6-, 9-, 12-, and 18-month follow-up), and the group-by-time interactions were added to the model. Furthermore, generalized linear models (GLM) was used to examine overall between-group treatment effect of the number
RkJQdWJsaXNoZXIy MjY0ODMw